English pronunciation error correction method and system based on artificial intelligence

Abstract

The invention discloses an English pronunciation error correction method and system based on artificial intelligence. The method comprises the following steps: S1, obtaining English pronunciation; s2, comparing the number of words; s3, after the recording information of the English reading fragments is formed again, the step S1 is executed; s4, marking a word sequence; s5, obtaining word pronunciation; s6, performing pronunciation comparison; s7, processing a comparison result; and S8, resetting the current word. According to the method, the recorded recording information is automatically compared with the English text, the comparison of the word number and the word pronunciation is included, and repeated learning or subsequent learning is carried out according to the comparison result. The system is suitable for the self-learning of English learners, and is convenient for the learners to grasp the learning effect and the learning progress by themselves.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence applications, and in particular, to an English pronunciation correction method and system based on artificial intelligence. Background Art

[0002] In the process of English learning, reading English aloud is an important process. For learners, when studying alone, it is often difficult to determine whether there are mistakes in their English reading or whether they meet the learning requirements. Usually, it is judged based on the reading model and personal feelings, and an objective judgment of the learning effect cannot be obtained. Summary of the Invention

[0003] To meet the above requirements, the present invention aims to provide an English pronunciation correction method and system based on artificial intelligence, which automatically compares the recorded audio information with the English text, including the comparison of the number of words and word pronunciations, and conducts repeated learning or subsequent learning according to the comparison results. It is suitable for the autonomous learning of English learners and facilitates learners to independently master the learning effect and learning progress.

[0004] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0005] First, the present invention discloses an English pronunciation correction method based on artificial intelligence, including the following steps:

[0006] S1. Obtain English pronunciation, obtain the audio information of the English reading fragment, and use the Viterbi algorithm to obtain the word sequence of the English reading fragment. The English reading fragment is matched with a corresponding English text;

[0007] S2. Compare the number of words, compare the number of words in the word sequence with the number of words in the corresponding English text. When the number of words in the word sequence is the same as the number of words in the corresponding English text, execute step S4;

[0008] When the number of words in the word sequence is different from the number of words in the corresponding English text, execute step S3;

[0009] S3. After reforming the audio information of the English reading fragment, execute step S1;

[0010] S4. Mark the word sequence, mark the word sequence in an incremental manner, and set the first word of the word sequence as the current word;

[0011] S5. Obtain word pronunciation, extract the characteristic information of the current word pronunciation, and obtain the characteristic information of the standard pronunciation of the word from the standard pronunciation library according to the word in the English text at the corresponding position of the current word;

[0012] S6. Pronunciation comparison: Compare the feature information of the pronunciation of the current word with the feature information of the standard pronunciation to obtain the similarity between the word pronunciation and the word standard pronunciation;

[0013] S7. Comparison result processing: When the similarity is not less than the predetermined value, execute step S7; when the similarity is less than the predetermined value, execute step S3;

[0014] S8. Reset the current word: Set the next word in the word sequence as the current word, and execute step S5; until the current word is the last word in the word sequence.

[0015] Preferably, in step S1, the Viterbi algorithm is as follows:

[0016] Use the words in the corresponding English text as the state space S of the hidden Markov model, with a total of k states, and the probability of the initial state i being π i , and the transition probability from state i to state j is α i'，j , and the output word sequence is y1,..., y T , and the most likely state sequence x1,..., x that generates the output result T is given by the recurrence relation:

[0017] V 1，k = P(y1|k)·π k

[0018] V t，k = max x∈S (P(y t |k)·a x，k ·V t-1,x )

[0019] where V t，k is the probability of the most likely corresponding state sequence of the observation results of the first t final states being k; the Viterbi path can be obtained by saving the backward pointer to remember the state x used in the above second equation. Declare a function Ptr(k,t), and the function Ptr(k,t) returns the x value used to calculate V t，k when t>1 or k when t = 1, where:

[0020] x T = arg max x∈S (V T，x )

[0021] x t-1 = Ptr(x t , t).

[0022] Preferably, a deep learning classifier is used to extract the feature information of the pronunciation of the current word.

[0023] Further preferably, the deep learning model adopted by the deep learning classifier is a fully connected neural network model.

[0024] Furthermore, in step S7, the predetermined value can be set.

[0025] Secondly, the present invention also discloses a system applicable to the above-mentioned English pronunciation correction method based on artificial intelligence, including:

[0026] A recording device for recording the English fragments read by the learner, where the English fragments are sentences or paragraphs;

[0027] A processing device for executing steps S1, S2, S4, S5, S6, S7, S8 and providing support modules for executing the above steps;

[0028] The processing device is connected to the recording device.

[0029] Furthermore, the processing device includes a logical operation module for executing steps S1, S2, S4, S5, S6, S7, S8.

[0030] Furthermore, the support modules include:

[0031] An input module for reading the recording information through the recording device;

[0032] A comparison module for comparing the number of words in the word sequence with that in the corresponding English text;

[0033] A word acquisition module for acquiring the word sequence of the English reading fragment;

[0034] A deep learning classifier module for extracting the feature information of the pronunciation of the current word;

[0035] A comparison module for comparing the feature information of the pronunciation of the current word with the feature information of the standard pronunciation;

[0036] A storage module for storing the corresponding English text;

[0037] A counter for setting the current word.

[0038] The learner reads and records the English text, and automatically compares the recorded information with the English text, including the comparison of the number of words and the pronunciation of words, and conducts repeated learning or subsequent learning according to the comparison results.

[0039] Advantages of the present invention:

[0040] 1. The present invention is applicable to the autonomous learning of English learners, facilitating the learners to master the learning effect and progress by themselves.

[0041] 2. The present invention can be applicable to English learners with different needs by setting a predetermined value of similarity and setting requirements for learning effects.

[0042] 3. The present invention uses artificial intelligence to obtain word sequences, with a high degree of intelligence and can adapt to different speech rates and other situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flowchart of Embodiment 1.

[0044] Figure 2 It is a principle block diagram of Embodiment 2. DETAILED DESCRIPTION OF THE INVENTION

[0045] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings.

[0046] Embodiment 1

[0047] This embodiment discloses an English pronunciation correction method based on artificial intelligence, as Figure 1 shown, which includes the following steps:

[0048] S1. Obtain English pronunciation, obtain the recording information of the English reading fragment, and use the Viterbi algorithm to obtain the word sequence of the English reading fragment. The English reading fragment is matched with the corresponding English text; the learner reads the English text, which can be a relatively casual reading of a whole sentence or a whole paragraph, etc., and records it to obtain the recording information of the English reading fragment. According to the obtained recording information of the English reading fragment, the Viterbi algorithm is used to obtain the word sequence of the English reading fragment, where:

[0049] The Viterbi algorithm is as follows:

[0050] Taking the words in the corresponding English text as the state space S of the hidden Markov model, there are a total of k states, and the probability of the initial state i is π i , and the transition probability from state i to state j is α i'，j , the output word sequence is y1,..., y T , and the most likely state sequence x1,..., x T that produces the output result is given by the recurrence relation:

[0051] V 1，k = P(y1|k)·π k

[0052] V t，k = max x∈S (P(y t |k)·a x,k ·V t-1,x )

[0053] where V t，k is the probability that the state sequence most likely corresponding to the first t observation results with the final state of k; the Viterbi path can be obtained by remembering the state x used in the second equation above by saving the backward pointer, declare a function Ptr(k,t), and the function Ptr(k,t) returns the x value used when calculating V t，k if t > 1 or k if t = 1, where:

[0054] x T = arg max x∈S (V T，x )

[0055] x t-1 = Ptr(x t ,t).

[0056] S2. Compare the number of words. Compare the number of words in the word sequence with the number of words in the corresponding English text. When the number of words in the word sequence is the same as the number of words in the corresponding English text, execute step S4;

[0057] When the number of words in the word sequence is different from the number of words in the corresponding English text, execute step S3;

[0058] S3. After reforming the recording information of the English reading segment, execute step S1;

[0059] where reforming the recording information of the English reading segment means that the learner rereads the English text containing the whole or part of the aforementioned English reading segment, and the learner can choose to read the whole or part by himself.

[0060] S4. Mark the word sequence, mark the word sequence in an increasing manner, and set the first word of the word sequence as the current word;

[0061] S5. Obtain the word pronunciation. Extract the feature information of the current word pronunciation. According to the word in the English text at the corresponding position of the current word, obtain the feature information of the standard pronunciation of the word from the standard pronunciation library;

[0062] The feature information of the current word pronunciation is extracted using a deep learning classifier.

[0063] where the deep learning model used by the deep learning classifier is a fully connected neural network model.

[0064] S6. Pronunciation comparison. Compare the feature information of the current word pronunciation with the feature information of the standard pronunciation to obtain the similarity between the word pronunciation and the word standard pronunciation;

[0065] S7. Comparison result processing: When the similarity is not less than the predetermined value, step S7 is executed; when the similarity is less than the predetermined value, step S3 is executed; wherein, the predetermined value is settable.

[0066] S8. Reset the current word, set the next word in the word sequence as the current word, and execute step S5; until the current word is the last word in the word sequence.

[0067] Thus, the learner has completed the learning of this English segment. Then subsequent learning can be carried out, and the same steps S1 to S8 are adopted for subsequent learning.

[0068] Embodiment 2

[0069] Based on Embodiment 1, this embodiment discloses a system applicable to the above-mentioned English pronunciation correction method based on artificial intelligence. Specifically, as Figure 2 shown, the system includes:

[0070] A recording device for recording the English segment read by the learner, where the English segment is a sentence or a paragraph;

[0071] A processing device for executing steps S1, S2, S4, S5, S6, S7, and S8, and providing support modules for executing the above steps;

[0072] The processing device is connected to the recording device.

[0073] The processing device includes a logical operation module, and the logical operation module is used to execute steps S1, S2, S4, S5, S6, S7, and S8.

[0074] The support modules include:

[0075] An input module for reading the recording information through the recording device;

[0076] A comparison module for comparing the number of words in the word sequence with the number of words in the corresponding English text;

[0077] A word acquisition module for acquiring the word sequence of the English reading segment;

[0078] A deep learning classifier module for extracting the feature information of the pronunciation of the current word;

[0079] A comparison module for comparing the feature information of the pronunciation of the current word with the feature information of the standard pronunciation;

[0080] A storage module for storing the corresponding English text;

[0081] A counter for setting the current word.

[0082] Of course, the present invention may have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art can make various corresponding changes and modifications according to the present invention. However, these corresponding changes and modifications shall fall within the protection scope of the appended claims of the present invention.

Claims

1. An English pronunciation error correction method based on artificial intelligence, characterized in that, It includes the following steps: S1. Obtain the English pronunciation, obtain the recording information of the English reading segment, and use the Viterbi algorithm to obtain the word sequence of the English reading segment. The English reading segment is matched with the corresponding English text; S2. Compare the number of words. Compare the number of words in the word sequence with the number of words in the corresponding English text. When the number of words in the word sequence is the same as the number of words in the corresponding English text, execute step S4; When the number of words in the word sequence is different from the number of words in the corresponding English text, execute step S3; S3. After reforming the recording information of the English reading segment, execute step S1; S4. Mark the word sequence, mark the word sequence in an increasing manner, and set the first word of the word sequence as the current word; S5. Obtain the word pronunciation, extract the characteristic information of the current word pronunciation, and obtain the characteristic information of the standard pronunciation of the word from the standard pronunciation library according to the word in the English text at the corresponding position of the current word; S6. Compare the pronunciations. Compare the characteristic information of the current word pronunciation with the characteristic information of the standard pronunciation to obtain the similarity between the word pronunciation and the word standard pronunciation; S7. Process the comparison result. When the similarity is not less than the predetermined value, execute step S7; When the similarity is less than the predetermined value, execute step S3; S8. Reset the current word, set the next word in the word sequence as the current word, and execute step S5; Until the current word is the last word of the word sequence.

2. The English pronunciation error correction method based on artificial intelligence according to claim 1, characterized in that, In step S1, the Viterbi algorithm is as follows: Let the words in the corresponding English text be the state space \(S\) of the hidden Markov model, with a total of \(k\) states. The probability of the initial state \(i\) is \(\pi\). i The transition probability from state \(i\) to state \(j\) is \(\alpha\). i'，j The output word sequence is \(y_1,\ldots,y\). T The most likely state sequence \(x_1,\ldots,x\) that generates the output result is given by the recurrence relation: T is given by the recurrence relation: V 1，k = P(y1|k)·π k V t，k = max x∈S (P(y t |k)·a x，k ·V t-1,x ) where V t，k is the probability of the state sequence most likely corresponding to the first t observations with final state k; the Viterbi path can be obtained by saving the back pointer to remember the state x used in the second equation above. Declare a function Ptr(k, t), where the function Ptr(k, t) returns the value of x used to calculate V t，k if t > 1 or k if t = 1, where: x T = arg max x∈S (V T，x ) x t-1 = Ptr(x t , t).

3. The English pronunciation error correction method based on artificial intelligence according to claim 1, wherein In step S5, a deep learning classifier is used to extract the characteristic information of the current word pronunciation.

4. The English pronunciation correction method based on artificial intelligence according to claim 3, wherein, The deep learning model adopted by the deep learning classifier is a fully connected neural network model.

5. The English pronunciation correction method based on artificial intelligence according to claim 1, characterized in that, In step S7, the predetermined value can be set.

6. A system applicable to the artificial intelligence-based English pronunciation correction method according to any one of claims 1-5, characterized in that It includes: A recording device for recording the learner's reading of an English segment, where the English segment is a sentence or a paragraph; A processing device for executing steps S1, S2, S4, S5, S6, S7, S8, and providing support modules for executing the above steps; The processing device is connected to the recording device.

7. The system according to claim 6, wherein The processing device includes a logical operation module for executing steps S1, S2, S4, S5, S6, S7, S8.

8. The system according to claim 6, wherein The support modules include: An input module for reading the recording information through the recording device; A comparison module for comparing the number of words in the word sequence with the number of words in the corresponding English text; A word acquisition module for acquiring the word sequence of the English reading segment; A deep learning classifier module for extracting the characteristic information of the current word pronunciation; A comparison module for comparing the characteristic information of the current word pronunciation with the characteristic information of the standard pronunciation; A storage module for storing the corresponding English text; A counter for setting the current word.

Images